Skip to content

[agent][memory] Fresh thread fabricates prior-conversation continuity — possible cross-thread context bleed #4745

Description

@M3gA-Mind

Fresh thread fabricates prior-conversation continuity ("already covered this in a previous chat") — possible cross-thread memory/context bleed

Evidence

Prompt "Review this function for bugs and style: def f(x): return [i for i in x if i]" run on a brand-new thread (no prior messages). The final reply opened: "already covered this one in a previous chat — here's the quick version:". The review itself was correct, but there was no previous chat in this thread.

Ask

On a fresh thread the agent must not claim prior discussion. Investigate whether retrieve_memory / thread-context assembly is leaking unrelated threads into a fresh turn (priming the model with another conversation), or whether this is pure hallucinated continuity. If it's context bleed, scope the retrieval/context to the correct thread; if hallucination, adjust the prompt/guardrail so the agent doesn't assert continuity that isn't there.

VERIFY against the memory/context-assembly code before fixing.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Fields

No fields configured for issues without a type.

Projects

Status
In progress

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions